Real-Time Coastal Flood Prediction Using Machine Learning Approach
Abstract
Among the natural disasters, flood is the most destructive event that causes massive damage to human life, critical infrastructure, and the environment. Real-time flood prediction is crucial to mitigate potential impacts of flooding on communities. This study examined machine learning algorithm (Random Forest) to predict flood risks in real-time across a coastal environmental setting in the Carolinas. The model is trained and evaluated using recent hurricane driven floods with various design periods. Analysis suggested that Random Forest performed well at predicting sharp and high peak rate flood hydrograph with a lower false negative rate. The results indicate that rainfall amount and spatial distribution as well as tidal surge were by far the most dominant input variables in flood severity prediction. Our approaches serve as a first step toward building model-driven methods to assess hurricane driven flooding occurrences in real-time across coastal environmental systems.
- Publication:
-
AGU Fall Meeting Abstracts
- Pub Date:
- December 2019
- Bibcode:
- 2019AGUFM.H33K2099S
- Keywords:
-
- 3305 Climate change and variability;
- ATMOSPHERIC PROCESSES;
- 1816 Estimation and forecasting;
- HYDROLOGY;
- 1817 Extreme events;
- HYDROLOGY;
- 4333 Disaster risk analysis and assessment;
- NATURAL HAZARDS